Attentional Encoder Network for Targeted Sentiment Classification

25 Feb 2019  ·  Youwei Song, Jiahai Wang, Tao Jiang, Zhiyue Liu, Yanghui Rao ·

Targeted sentiment classification aims at determining the sentimental tendency towards specific targets. Most of the previous approaches model context and target words with RNN and attention. However, RNNs are difficult to parallelize and truncated backpropagation through time brings difficulty in remembering long-term patterns. To address this issue, this paper proposes an Attentional Encoder Network (AEN) which eschews recurrence and employs attention based encoders for the modeling between context and target. We raise the label unreliability issue and introduce label smoothing regularization. We also apply pre-trained BERT to this task and obtain new state-of-the-art results. Experiments and analysis demonstrate the effectiveness and lightweight of our model.

PDF Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Aspect-Based Sentiment Analysis (ABSA) SemEval-2014 Task-4 AEN-GloVe Restaurant (Acc) 80.98 # 31
Laptop (Acc) 73.51 # 32
Mean Acc (Restaurant + Laptop) 77.25 # 30
Aspect-Based Sentiment Analysis (ABSA) SemEval-2014 Task-4 AEN-BERT Restaurant (Acc) 83.12 # 17
Laptop (Acc) 79.93 # 11
Mean Acc (Restaurant + Laptop) 81.53 # 14
Aspect-Based Sentiment Analysis (ABSA) SemEval-2014 Task-4 BERT-SPC Restaurant (Acc) 84.46 # 14
Laptop (Acc) 78.99 # 14
Mean Acc (Restaurant + Laptop) 81.73 # 13
Sentiment Analysis Twitter AEN-BERT Accuracy 74.71 # 1
Sentiment Analysis Twitter AEN-GloVe Accuracy 72.83 # 3
Sentiment Analysis Twitter BERT-SPC Accuracy 73.55 # 2

Methods